How to Tell if an Image Is AI-Generated: 9 Signs to Check
AI image generators like Midjourney and DALL-E leave telltale artifacts. Here are the nine signs to check before you trust — or share — a photo.
AI image generators have crossed the line where casual inspection fails. Studies keep showing the same thing: people believe they can spot AI images, and then score barely better than a coin flip. The checks below give you a repeatable process instead of a gut feeling.
Quick checklist
- Zoom into the details. Skin that looks airbrushed everywhere, hair strands that merge into skin or clothing, earrings that don’t match, glasses that blend into eyebrows.
- Read any text in the image. Signs, labels, jerseys, and shopfronts are still a weak point — letters warp, duplicate, or turn into alphabet-shaped noise.
- Check lighting and shadows. Do all shadows fall in the same direction? Do reflections in mirrors, windows, or eyes match the scene? Generators compose scenes statistically, not physically.
- Look at backgrounds. Crowds with melted faces, buildings whose windows don’t line up, fences and railings that change pattern mid-span.
- Count and compare. Fingers are better than they were, but AI still fumbles repetition: rows of teeth, buttons, spokes, tiles, and window panes often drift.
- Examine symmetry. Pupils of different shapes, mismatched ears, collars and necklaces that don’t continue on both sides.
- Check texture transitions. Where hair meets forehead, where a hand grips an object, where clothing meets skin — these boundaries are where models blur or smear.
- Reverse image search it. Google Lens and TinEye show you where the image appeared first. A “breaking news photo” with no news source behind it is a red flag.
- Interrogate the context. Who posted it? Is the account new? Does the caption push an emotional reaction? Fabricated images spread on outrage, not accuracy.
Why manual checks are getting harder
The classic artifacts — six fingers, garbled text, plastic skin — were symptoms of early diffusion models. Each model generation cleans them up. What persists longer is physics: consistent lighting, coherent reflections, and plausible geometry are genuinely hard, because the model has no 3D scene, only patterns learned from pixels.
That’s also why no single tell is proof. A real photo can have weird lighting; an AI image can have perfect hands. You’re building a case, not finding a smoking gun.
When you need more than your eyes
Manual inspection breaks down on small, compressed, or cropped images — which is exactly what you see in feeds and chats. Automated detectors analyze statistical fingerprints left by the generation process itself, which survive even when the visible artifacts don’t.
Frequently asked questions
Can AI-generated images be detected with 100% accuracy?
No. Detection — human or automated — is probabilistic. Modern detectors report a likelihood score rather than a guarantee, and the best results come from combining automated analysis with the manual checks in this guide.
Do AI images still have messed-up hands in 2026?
Much less than they used to. Hands, teeth, and text were the classic giveaways, but recent diffusion models handle them far better. Treat clean hands as no evidence either way, and rely on lighting, physics, and context checks instead.
Does reverse image search prove a photo is real?
It can't prove authenticity, but it's a strong signal. If a photo claims to show a news event and no reputable outlet carries it, or the earliest copy appears on an anonymous account, be suspicious.